Baby hip joint standard body position ultrasonic imaging probe self-adaptive adjustment robot system

Abstract

The invention discloses a baby ultrasonic hip joint imaging standard body position probe self-adaption robot system which comprises an ultrasonic probe module, a mechanical arm artificial intelligence module, an image automatic recognition module and an image analysis module. The mechanical arm artificial intelligence module is connected with the image automatic identification module and the image analysis module, and the ultrasonic probe module and the image automatic identification module are respectively connected with the image analysis module. The system can automatically control the ultrasonic probe to conduct ultrasonic detection on the hip joint of the infant, the ultrasonic probe can adjust the position, the angle and the stress damping in a self-adaptive mode relative to the detection part, the detection precision is improved, ultrasonic images can be automatically analyzed, and a preliminary diagnosis result is obtained.

Description

Technical Field

[0001] The invention belongs to the field of ultrasonic tissue detection, and in particular relates to a robot system for adaptively adjusting an ultrasonic imaging probe of a standard position of an infant hip joint. Background Art

[0002] Developmental dysplasia of the hip (DDH) is a developmental disease of infants and young children that requires early diagnosis and treatment. The Graf method is one of the main diagnostic methods for DDH. To obtain a standard Graf hip joint image, eight anatomical landmarks (bone-cartilage junction, femoral head, synovial folds, joint capsule, acetabular labrum, bone dome, cartilage dome, and lower edge of the ilium) and three marking points on the standard plane (acetabular labrum, lower edge of the ilium, and middle section) must be met. Through the three marking points, three lines (baseline, bone dome line, and cartilage dome line) are obtained, and the angle between the baseline and the bone dome line (ɑ angle) and the angle between the baseline and the bone dome line (β angle) are measured. The ɑ angle and β angle are combined with the infant's age for classification and diagnosis.

[0003] The Graf method is to place the baby on his side on the examination bed, with one doctor keeping the hip to be examined slightly internally rotated and flexed into a standard position. The guide device of the ultrasound examination platform fixes the probe, and the robot arm moves the probe back and forth to the part of the baby to be examined. Another doctor holds the probe and presses it at a certain angle on the greater trochanter of the hip. The captured image is generated based on the echo, and the position and angle of the probe emitting sound waves are continuously adjusted to obtain a satisfactory standard coronal section of the hip joint. The current platform robot arm has a low degree of intelligence and requires manual control of the probe movement.

[0004] With the increasing maturity of robotics and AI technologies, in terms of AI automatic image acquisition technology, a 6-degree-of-freedom robot-assisted medical imaging method has been used for breast ultrasound diagnosis, which can automatically acquire ultrasound images instead of manual operations. However, in terms of automatic acquisition of hip ultrasound images, research on acquiring ultrasound images with the help of robotic technology is still in its infancy, and it faces many technical problems that need to be solved. For example, babies struggle and move easily when restrained, and their random movements will interfere with the imaging quality of ultrasound images, the accuracy of standard sections is low, and ultrasound images need to be analyzed manually and cannot be automatically analyzed. Summary of the invention

[0005] In order to solve the problems in the above-mentioned prior art that it is impossible to automatically acquire ultrasound images with the help of robot technology and that the random movements of infants interfere with the imaging quality of ultrasound images, the present application proposes a robot system for adaptively adjusting the ultrasound imaging probe of standard infant hip joint position. The robot controls the ultrasound probe to automatically reach the detection site through path planning, automatically captures the hip joint image, and automatically analyzes the image to draw auxiliary diagnostic conclusions. At the same time, the probe is controlled to adaptively adjust the random stress of the infant during the detection process and the detection accuracy.

[0006] The specific application is as follows:

[0007] An adaptive robot system for standard position probe of infant ultrasonic hip joint imaging, comprising an ultrasonic probe module, a mechanical arm artificial intelligence module, an automatic image recognition module and an image analysis module, wherein the ultrasonic probe module is connected to the mechanical arm artificial intelligence module and the automatic image recognition module, the mechanical arm artificial intelligence module is connected to the automatic image recognition module and the image analysis module, and the ultrasonic probe module and the automatic image recognition module are respectively connected to the image analysis module;

[0008] The ultrasound probe module is responsible for controlling the probe to capture ultrasound images of the infant's hip joint, including controlling the emission and pause of corresponding ultrasound waves according to preset values, and is responsible for acquiring probe images in real time through sensors or cameras;

[0009] The robotic arm artificial intelligence module is responsible for controlling the robotic arm, which further controls the precise movement, positioning and damping of random stress of the probe;

[0010] The automatic image recognition module uses machine learning technology to automatically recognize ultrasound sections that meet medical requirements in ultrasound images, thereby reducing the manual operation of doctors.

[0011] The image analysis module applies image recognition technology to simulate the expert diagnosis process and realize intelligent image analysis and diagnosis.

[0012] Furthermore, the image automatic recognition module includes an image collection submodule, an image preprocessing submodule, a machine modeling submodule and a model optimization submodule, wherein the image collection submodule is connected to the image preprocessing submodule, the image preprocessing submodule is connected to the machine modeling submodule, and the machine modeling submodule is connected to the model optimization submodule;

[0013] The image collection submodule is used to obtain a large number of infant hip joint ultrasound images. These images should include hip joint images of different angles and different infants. Professional doctors mark the key features of the hip joint on the images, including 8 anatomical landmarks and 3 identification points.

[0014] The image preprocessing submodule is used to remove noise and artifacts from the image, improve image quality, adjust the size and contrast of the image to make it suitable for subsequent processing, and enhance the key features annotated in the image;

[0015] The machine modeling submodule uses a convolutional neural network for model training, including using the convolutional neural network to automatically extract key features in the image, dividing the historical ultrasound images into a training set and a test set, training the recognition prediction model with the training set, and tuning the model with the test set. The recognition prediction model is used to identify ultrasound sections containing 8 anatomical landmarks and 3 identification points;

[0016] The model optimization submodule defines the cross entropy loss function, which is used to measure the difference between the model prediction and the actual annotation. The SGD optimization algorithm is selected to train the network and find the minimum value of the loss function, thereby training a performance-optimized model.

[0017] Further, the robot arm artificial intelligence module also includes a path planning submodule, a motion control submodule, an angle control submodule, a tactile perception submodule, a coupling agent automatic replacement and cleaning submodule, a dynamic human-machine coupling submodule, an ultrasonic probe adaptive adjustment submodule and a user interaction submodule, wherein the path planning submodule is connected to the motion control submodule and the coupling agent automatic replacement and cleaning submodule, the motion control submodule is respectively connected to the tactile perception submodule, the angle control submodule, the dynamic human-machine coupling submodule and the ultrasonic probe adaptive adjustment submodule, and the path planning submodule, the motion control submodule, the coupling agent automatic replacement and cleaning submodule, the angle control submodule, the dynamic human-machine coupling submodule and the ultrasonic probe adaptive adjustment submodule are respectively connected to the user interaction submodule;

[0018] Path planning submodule: responsible for calculating the optimal path for the ultrasound probe to reach the baby's examination site, ensuring that the probe can reach the target position quickly and accurately, and controlling the start and stop of the robotic arm;

[0019] Motion control submodule: The motion planning algorithm is used to convert the path planning information provided by the path planning submodule, the dynamic path adjustment unit, the angle adjustment unit and the feature matching analysis unit into specific motion instructions for each joint of the robotic arm. The motion controller controls the movement and operation of the robotic arm to ensure that the ultrasound probe can move smoothly and accurately to the predetermined position according to the planned path.

[0020] Tactile perception submodule: Through the feedback of force tactile array sensor, proximity array sensor and joint torque sensor, the robot arm is controlled to simulate the doctor's arm, and the contact area and contact force between the probe and the baby's skin are adjusted to achieve the best optimization;

[0021] The angle control submodule uses a machine learning algorithm to predict the initial angle of the ultrasonic probe at the inspection position, and after receiving the position arrival instruction, adjusts the angle value of the probe at the inspection position to meet the angle setting requirements;

[0022] Coupling agent automatic replacement and cleaning submodule: realizes automatic replacement of disposable coupling agent before examination, and automatic cleaning and sterilization of the probe and examination site after examination;

[0023] Dynamic human-machine coupling submodule: adjusts the probe's resistance to the stress from the random movements of the baby, solving the problem of the baby's random movements interfering with the probe's imaging quality;

[0024] Ultrasonic probe adaptive adjustment submodule: According to the positional relationship between the detected image and the predicted standard section, the position and angle of the ultrasonic probe are automatically adjusted to accurately capture the standard section;

[0025] User interaction submodule: allows doctors or operators to interact with the robotic arm, including data input and necessary manual adjustments and interventions.

[0026] Further, the path planning submodule includes an environment perception unit, a map reconstruction unit, a target recognition and positioning unit, a path search algorithm unit, a dynamic path adjustment unit and a user intervention interface unit, wherein the environment perception unit is connected to the map reconstruction unit, the target recognition and positioning unit is connected to the path search algorithm unit, and the map reconstruction unit, the dynamic path adjustment unit and the user intervention interface unit are all connected to the path search algorithm unit;

[0027] The environment sensing unit collects information about the examination environment, including the position of the infant, the specific location of the infant's examination part, and the layout of the examination bed;

[0028] The map reconstruction unit constructs or updates a map of the inspection area for path planning based on data provided by the environment perception submodule;

[0029] The target recognition and positioning unit uses a convolutional neural network to recognize the specific location of the infant's examination site, including the greater trochanter of the femur at the hip, and uses a spatial transformation technology to convert the hip joint image coordinates into actual physical space coordinates to provide accurate examination site coordinate information, while obtaining the current position coordinates of the probe through a position detection sensor;

[0030] The path search algorithm unit uses a path search algorithm to calculate an optimal path from a current position to a target position, and the algorithm includes an A* algorithm, a Dijkstra algorithm or an RRT algorithm;

[0031] The dynamic path adjustment unit monitors environmental changes and infant movements in real time during path execution, and dynamically adjusts the path to adapt to new situations, including using sensors and cameras to monitor the environment and infant movements in real time, using computer vision technology to track the movement of the infant and the examination site, analyzing movement patterns and predicting future positions, and applying a path adjustment algorithm to recalculate or adjust the path;

[0032] The user intervention interface unit allows a doctor or operator to manually intervene in the path planning process when necessary to ensure safety and examination quality.

[0033] Further, the angle control submodule includes a machine learning unit, a probe orientation unit and an angle adjustment unit, and the machine learning unit and the probe orientation unit are respectively connected to the angle adjustment unit;

[0034] The machine learning unit collects historical infant hip joint detection data, obtains a three-dimensional angle value of the relative inspection position coordinates of the ultrasound probe through machine learning, and transmits the three-dimensional angle value to the angle adjustment unit; wherein the historical infant hip joint detection data includes infant age, physiological parameters, body position parameters, ultrasound parameters and coupling agent parameters;

[0035] The probe orientation unit receives the probe image, identifies the probe reference point and the specific part of the hip joint according to medical common sense, and draws a line between the two to determine the probe ultrasound emission direction;

[0036] The angle adjustment unit receives the predicted three-dimensional angle value, and then after receiving the in-place instruction issued by the motion control submodule, uses the image registration technology to integrate the probe image and the inspection position into the image comparison system in the same coordinate space, establishes a three-dimensional coordinate system with the inspection position coordinates as the origin, and uses the vector analysis method to measure whether the angles between the probe's ultrasonic emission direction and the inspection position's XYZ coordinates meet the three-dimensional angle value. If not, an angle adjustment instruction is generated and sent to the motion control submodule to adjust the probe angle.

[0037] Further, the dynamic human-machine coupling submodule includes a probe force detection unit, a data preprocessing unit, a controller unit and a stability analysis unit, the probe force detection unit is connected to the data preprocessing unit, the data preprocessing unit is connected to the controller unit, and the controller unit is connected to the stability analysis unit;

[0038] The probe force detection unit measures the random force signal applied to the probe in real time by integrating a sensor located at the probe;

[0039] The data preprocessing unit filters, amplifies and digitizes the signals collected by the sensor to extract useful information;

[0040] The controller unit is used to design a dynamic characteristic model that can describe the interaction between the probe and the infant, convert the force signal into a control signal, and design a control algorithm to dynamically adjust the damping of the probe against stress according to the random force signal applied to the probe;

[0041] The stability analysis unit uses control theory and Lyapunov method to analyze the stability of the controller.

[0042] Furthermore, the ultrasound probe adaptive adjustment submodule includes a key feature positioning unit, a standard section prediction unit, a probe angle and position measurement unit and a feature matching analysis unit, and the key feature positioning unit, the standard section prediction unit and the probe angle and position measurement unit are respectively connected to the feature matching analysis unit;

[0043] The key feature positioning unit is used to receive the ultrasound section image containing 8 anatomical landmarks and 3 identification points recognized in real time by the image recognition submodule, and locate the key features in the image using a convolutional neural network algorithm in the standard section image, the key features including the 8 anatomical landmarks and the 3 identification points;

[0044] The standard section prediction unit uses convolutional neural network training to obtain the standard section, calibrates the position coordinates of key features on the standard section in the XY coordinate system, and fits the connection lines between different key features. The standard section must include 8 anatomical landmarks and 3 identification points. The historical parameters collected during training include infant age, physiological parameters, body position parameters, ultrasound parameters, and coupling agent parameters. The model is tuned and the optimization algorithm is used to improve the accuracy of the model.

[0045] The probe angle and position measurement unit obtains the probe image in real time to measure the initial position and angle of the probe;

[0046] The feature matching and analysis unit acquires the real-time image and identifies multiple key features therein, and receives the standard section input by the user interaction submodule at the same time, integrates the trained standard section, real-time image and probe image into the image comparison system in the same coordinate space by using the image registration technology, acquires the same key features of the standard section and the real-time image, and compares and analyzes the positions and angles of the same key features of the above different images, calculates the difference between the position of the key feature in the real-time image and the position of the same key feature in the standard section, and then obtains the position error, calculates the difference between the angle of the probe and the angle of the standard section to obtain the angle error, and acquires the initial position and angle of the probe in the same coordinate space by using the image analysis technology according to the probe image, and sends the above position error, angle error and the initial position and angle information of the probe to the motion control submodule;

[0047] The motion control submodule is used to receive the control signal transmitted by the feature matching analysis unit, which contains angle error information, position error information, and the initial position and angle information of the probe. The target position and target angle information are calculated based on the angle error and position error information, and the initial position, initial angle, target position and target angle information are parsed into instructions for the movement of the robotic arm. The motion path of the robotic arm from the initial position to the target position and the rotation path from the initial angle to the target angle are planned, and the probe is driven from the current position to the target position and rotated by the corresponding angle through the robotic arm controller.

[0048] Furthermore, the feature matching analysis unit includes a position error calculation subunit and an angle error calculation subunit, wherein the position error calculation subunit obtains the coordinates (XY plane) of a preset key feature from the real-time image. 当前 , Y 当前 ), obtain the coordinates of the same key feature on the XY plane from the predicted standard section (X 标准 , Y 标准 ), using direct comparison method: ΔX=(X 标准 -X 当前 ), ΔY=(Y 标准 -Y 当前 ), where (ΔX, ΔY) constitutes the position error;

[0049] The angle error calculation subunit uses a feature detection algorithm to obtain the positions of two preset key features from the real-time image, uses a straight line detection algorithm to obtain a line between the two preset key features, and defines the line as a vector A, and determines the direction of the vector;

[0050] Using a feature detection algorithm to obtain the positions of the same two preset key features from the predicted standard section, using a straight line detection algorithm to obtain a line between the two preset key features, and defining the line as a vector B, and determining the direction of the vector;

[0051] Using the formula Calculate the angle θ between the two vectors A and B, where · represents the dot product and ∥·∥ represents the modulus of the vectors. The angle θ constitutes the angle error.

[0052] Further, the image analysis module includes an image recognition submodule, a key point positioning submodule, a line extraction submodule, an angle calculation submodule and a typing diagnosis submodule, the image recognition submodule is connected to the identification point positioning unit, the identification point positioning unit is connected to the line extraction submodule, the line extraction submodule is connected to the angle calculation submodule, and the angle calculation submodule is connected to the typing diagnosis submodule;

[0053] The image recognition submodule receives the standard section containing 8 anatomical landmarks and 3 identification points captured by the ultrasound probe adaptive adjustment submodule, and uses machine learning technology to identify the cartilage roof, acetabular bone roof, the lower edge of the ilium in the deep part of the acetabular fossa, the acetabular labrum and the middle section of the acetabular bone roof;

[0054] The key point positioning submodule uses a feature detection algorithm to enhance the edge and locate the position of the key point from the hip joint image identified by the image recognition submodule;

[0055] The line extraction submodule uses a line detection algorithm to extract a straight line from the image based on the middle section of the acetabulum top, the position of the key points and the straight line definition;

[0056] The angle calculation submodule converts the lines into vectors and uses trigonometric functions to calculate the angles between them, wherein the angle between the baseline and the bone top line is defined as angle α, and the baseline intersects with the cartilage top line to form angle β;

[0057] The typing diagnosis submodule performs threshold judgment on the above angles and outputs corresponding typing results, wherein the threshold typing includes Graf I type: α angle ≥ 60°; Graf II type: 50°≤α<60°; Graf IIc type: 43°≤a<50°, β≤77°; Graf D type: 43°≤a<50°, β>77°; Graf IIIa type: a<43°; Graf IIIb type: a<43°; Graf IV type: a<43°;

[0058] Wherein, the key point positioning submodule includes an edge enhancement subunit and a corner point detection subunit;

[0059] The edge enhancement subunit uses an edge detection algorithm to highlight the lower edge of the ilium, the acetabular labrum and the middle section of the acetabular bone top in the deep part of the acetabular fossa in the image;

[0060] The corner point detection subunit uses Harris corner point detection technology to identify key points in the edge of the image, and the key points include the top of the acetabulum, the lower edge of the ilium in the deep part of the acetabulum, the acetabulum bone edge and the midpoint of the acetabular labrum.

[0061] Further, the line extraction submodule includes a line definition unit, a candidate line detection unit and an optimal line determination unit;

[0062] The straight line definition unit is used to define a straight line, wherein the baseline is defined as a tangent line of the ilium along the transition from the proximal perichondrium to the periosteum; the bone top line is defined as a tangent line of the acetabular bone top from the lower edge of the ilium in the deep part of the acetabular fossa; the cartilage top line is defined as a line connecting the bone edge to the midpoint of the labrum;

[0063] The candidate line detection unit maps the edge points and key points of the interface in the middle of the ilium to the parameter space through Hough transform according to the definition of the line, and detects the line through the accumulator method to form a candidate line;

[0064] The best straight line determination unit uses the RANSAC algorithm to randomly select a group of points from the above edge points or key points for straight line fitting, and evaluates the consistency of other points with the straight line, and finally determines the best straight line.

[0065] The candidate line detection unit maps the edge points and key points of the middle interface of the acetabulum top to the parameter space through Hough transform according to the definition of the line, and detects the line through the accumulator method to form the candidate line;

[0066] The best straight line determination unit uses the RANSAC algorithm to randomly select a group of points from the above edge points or key points for straight line fitting, and evaluates the consistency of other points with the straight line, and finally determines the best straight line.

[0067] The beneficial effects of the infant hip joint standard position ultrasound imaging probe adaptive adjustment robot system of the present invention are as follows:

[0068] The present invention uses a program-controlled robot to plan the optimal path for the initial position of the ultrasound probe to reach the examination site of the infant. The motion control submodule shifts the ultrasound probe from the initial position to the greater trochanter of the femur at the infant's hip according to the planned path, and uses machine learning to predict the preset angle value of the probe at the greater trochanter of the femur at the infant's hip to ensure that an ultrasound section containing at least 8 anatomical landmarks and 3 identification points is obtained. The ultrasound section that meets the requirements of containing 8 anatomical landmarks and 3 identification points is identified through machine learning. During the capture process, the captured ultrasound section is compared with the predicted standard section image for key features to adaptively adjust the position and angle of the probe, so as to ensure that the captured ultrasound section is more accurately in line with the position and angle requirements of the standard section, and also realize the adaptive adjustment function when the angle and position of the probe change slightly;

[0069] The present invention adopts a controller unit to describe the dynamic characteristic model of the interaction between the probe and the baby, converts the force signal into a control signal, and designs a control algorithm to dynamically adjust the damping of the probe against the random stress of the baby according to the random force signal applied to the probe, so that the probe automatically resists the peristaltic stress of the baby during the image capture process, and avoids the probe changing position or angle under the stress of the baby, which affects the imaging quality;

[0070] The present invention uses an image analysis module to automatically obtain the baseline, bone top line and cartilage top line for standard sections that meet the requirements according to medical definitions, and automatically calculates the α angle and β angle through software technology. The Graf diagnostic method is used to automatically make a preliminary typing diagnosis based on the values ​​of the α angle and the β angle. The typing results provide a reference for doctors to further diagnose and treat clinically, thereby reducing the labor intensity of manual diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 This is a schematic diagram of the robot system for adaptively adjusting the ultrasound imaging probe for standard position of infant hip joints according to the present invention;

[0072] Figure 2 This is an architecture diagram of the path planning submodule of the present invention;

[0073] Figure 3 This is a structural diagram of the angle control submodule of the present invention;

[0074] Figure 4 This is a schematic diagram of the dynamic human-machine coupling submodule of the present invention;

[0075] Figure 5 This is a schematic diagram of the ultrasonic probe adaptive adjustment submodule of the present invention;

[0076] Figure 6 This is a schematic diagram of the line extraction submodule of the present invention;

[0077] Figure 7 This is an architecture diagram of the key point positioning submodule of the present invention;

[0078] Figure 8 This is a schematic diagram of the line extraction submodule of the present invention;

[0079] Fig. 9 It is a structural diagram of the feature matching analysis unit of the present invention. DETAILED DESCRIPTION

[0080] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0081] An adaptive robot system for standard position probe of infant ultrasonic hip joint imaging, comprising an ultrasonic probe module, a mechanical arm artificial intelligence module, an automatic image recognition module and an image analysis module, wherein the ultrasonic probe module is connected to the mechanical arm artificial intelligence module and the automatic image recognition module, the mechanical arm artificial intelligence module is connected to the automatic image recognition module and the image analysis module, and the ultrasonic probe module and the automatic image recognition module are respectively connected to the image analysis module;

[0082] The ultrasound probe module is responsible for controlling the probe to capture the ultrasound image of the infant's hip joint, including controlling the emission and pause of the corresponding ultrasound according to preset values, and is responsible for acquiring the probe image in real time through a sensor or camera, and then transmitting the captured probe image to the robotic arm artificial intelligence module and the image analysis module.

[0083] The robotic arm artificial intelligence module is responsible for controlling the robotic arm, and further controlling the precise movement, positioning and damping of random stress of the probe through the robotic arm; the robotic arm can control the ultrasound probe to adjust the position and angle, and move the ultrasound probe from the initial position to the detection position according to the path planning, which is generally the greater trochanter of the femur at the hip. The robotic arm can control the probe to apply pressure to the hip joint to resist the corresponding random stress of the baby, thereby preventing the probe from slightly changing its angle or position under the action of random stress, which would affect the accuracy of image capture.

[0084] The image automatic recognition module uses machine learning technology to automatically identify the ultrasonic sections in the ultrasound image that meet the medical requirements, reducing the doctor's manual operation. Traditionally, the doctor holds the ultrasound probe to capture the image and constantly fine-tunes the angle of the ultrasound probe to identify the ultrasound image with the echo of 8 anatomical landmarks. It is also necessary to make manual judgments on whether the ultrasound image further meets the standard section requirements, which is relatively cumbersome. This application automatically recognizes ultrasound images with 8 anatomical landmark echoes, and combines machine learning and influencing medical knowledge to further judge the similarity between the ultrasound image and the standard section. By fine-tuning the probe, the most similar ultrasound image is found for capture, with a high degree of automation.

[0085] The image analysis module applies image recognition technology to simulate the expert diagnosis process, realizes intelligent image analysis and diagnosis, avoids the traditional steps of manually identifying medical anatomical landmarks on standard sections and manually drawing lines and measuring angles, and provides a basis for further disease diagnosis.

[0086] In specific implementation, the image automatic recognition module includes an image collection submodule, an image preprocessing submodule, a machine modeling submodule and a model optimization submodule, wherein the image collection submodule is connected to the image preprocessing submodule, the image preprocessing submodule is connected to the machine modeling submodule, and the machine modeling submodule is connected to the model optimization submodule;

[0087] The image collection submodule is used to obtain a large number of infant hip joint ultrasound images. These images should include hip joint images of various angles and different infants, preferably hip joint ultrasound images of infants of similar or same age. Key features of the hip joint, including 8 anatomical landmarks and 3 identification points, should be marked on the images by professional doctors.

[0088] The image preprocessing submodule is used to remove noise and artifacts from the image, improve image quality, adjust the size and contrast of the image to make it suitable for subsequent processing, and enhance the key features annotated in the image;

[0089] The machine modeling submodule uses a convolutional neural network for model training, including using the convolutional neural network to automatically extract key features in the image, dividing the historical ultrasound images into a training set and a test set, training the recognition prediction model with the training set, and tuning the model with the test set. The recognition prediction model is used to identify ultrasound sections containing 8 anatomical landmarks and 3 identification points;

[0090] In specific implementation, the model optimization submodule is used to optimize the model, including defining the cross entropy loss function to measure the difference between the model prediction and the actual annotation, selecting the SGD optimization algorithm to train the network, and finding the minimum value of the loss function, so as to train a model with better performance; it also includes using the cross-validation method to evaluate the generalization ability of the model, evaluating the accuracy, recall rate and F1 score indicators of the model on an independent test set; and adjusting the network's hyperparameters, including the learning rate and batch size, to optimize the model performance; increasing the diversity of training data through data enhancement technology to improve the robustness of the model.

[0091] In specific implementation, the robotic arm artificial intelligence module also includes a path planning submodule, a motion control submodule, an angle control submodule, a tactile perception submodule, an automatic replacement and cleaning submodule of a coupling agent, a dynamic human-machine coupling submodule, an ultrasonic probe adaptive adjustment submodule and a user interaction submodule, wherein the path planning submodule is connected to the motion control submodule and the automatic replacement and cleaning submodule of the coupling agent, the motion control submodule is respectively connected to the tactile perception submodule, the angle control submodule, the dynamic human-machine coupling submodule and the ultrasonic probe adaptive adjustment submodule, and the path planning submodule, the motion control submodule, the automatic replacement and cleaning submodule of the coupling agent, the angle control submodule, the dynamic human-machine coupling submodule and the ultrasonic probe adaptive adjustment submodule are respectively connected to the user interaction submodule.

[0092] Path planning submodule: responsible for calculating the optimal path for the ultrasound probe to reach the baby's examination area, ensuring that the probe can reach the target position quickly and accurately, and controlling the start and stop of the robotic arm.

[0093] Motion control submodule: Use the motion planning algorithm to convert the path planning information provided by the path planning submodule, dynamic path adjustment unit, angle adjustment unit and feature matching analysis unit into specific motion instructions for each joint of the robotic arm. The movement and operation of the robotic arm are controlled by the motion controller. The position, speed and force feedback of the robotic arm joints are provided by feedback sensors such as the tactile perception submodule, and adaptive force and other adjustments are made. The motion state of the robotic arm is monitored in real time to ensure that the ultrasound probe can move smoothly and accurately to the predetermined position according to the planned path.

[0094] Tactile perception submodule: Receives the arrival signal from the path planning submodule, and then through the feedback of the force tactile array sensor, proximity array sensor and joint torque sensor, the robotic arm simulates the doctor's arm, and the motion control submodule adjusts the contact area and contact force between the probe and the baby's skin to achieve the best result, avoiding excessive pressure on the probe and causing discomfort to the child. The detection and adjustment process continues until the standard section is captured. When the stress adjustment conflicts with the resistance damping, the instructions of the dynamic human-machine coupling submodule are executed.

[0095] The angle control submodule uses a machine learning algorithm to predict the initial angle prediction value of the ultrasonic probe at the inspection position, and after receiving the position arrival instruction from the path planning submodule, adjusts the angle value of the probe at the inspection position to meet the angle requirement.

[0096] Automatic coupling agent replacement and cleaning submodule: It realizes automatic replacement of disposable coupling agent before the motion control submodule starts running, and automatic cleaning and sterilization of the probe and the inspection part after the inspection. For example, after identifying and grasping the standard section, it receives the completion instruction of the probe adaptive adjustment submodule and starts cleaning and sterilization.

[0097] Dynamic human-machine coupling submodule: After receiving the arrival signal from the path planning submodule, it adjusts the probe's resistance to the stress from the random movements of the baby, solving the problem of the baby's random movements interfering with the imaging quality of the probe.

[0098] Ultrasound probe adaptive adjustment submodule: According to the positional relationship between the detected image and the predicted standard section, the position and angle of the ultrasound probe are automatically adjusted to accurately capture the standard section.

[0099] User interaction submodule: allows the doctor or operator to interact with the robotic arm, and is connected to the rest of the above modules, including data input and necessary manual adjustments or interventions.

[0100] In a specific implementation, the path planning submodule includes an environment perception unit, a map reconstruction unit, a target recognition and positioning unit, a path search algorithm unit, a dynamic path adjustment unit, and a user intervention interface unit, wherein the environment perception unit is connected to the map reconstruction unit, the target recognition and positioning unit is connected to the path search algorithm unit, and the map reconstruction unit, the dynamic path adjustment unit, and the user intervention interface unit are all connected to the path search algorithm unit;

[0101] The environment sensing unit collects information about the examination environment through a camera, etc., including the position of the baby, the specific position of the baby's examination part, and the layout of the examination bed;

[0102] The map reconstruction unit constructs or updates a map of the inspection area for path planning based on data provided by the environment perception submodule;

[0103] The target recognition and positioning unit uses a convolutional neural network to recognize the specific location of the infant's examination site, including the greater trochanter of the femur at the hip, and uses a spatial transformation technology to convert the hip joint image coordinates into actual physical space coordinates to provide accurate examination site coordinate information, while obtaining the current position coordinates of the probe through a position detection sensor;

[0104] The path search algorithm unit uses a path search algorithm to calculate an optimal path from a current position to a target position, and the algorithm includes an A* algorithm, a Dijkstra algorithm or an RRT algorithm;

[0105] The dynamic path adjustment unit monitors environmental changes and infant movements in real time during path execution, and dynamically adjusts the path to adapt to new situations, including using sensors and cameras to monitor the environment and infant movements in real time, using computer vision technology to track the movement of the infant and the examination site, analyzing movement patterns and predicting future positions, and applying a path adjustment algorithm to recalculate or adjust the path;

[0106] The user intervention interface unit allows a doctor or operator to manually intervene in the path planning process when necessary to ensure safety and examination quality;

[0107] Through the above steps, the probe is automatically moved to the preset inspection position coordinates.

[0108] In a specific implementation, the angle control submodule includes a machine learning unit, a probe orientation unit and an angle adjustment unit, and the machine learning unit and the probe orientation unit are respectively connected to the angle adjustment unit;

[0109] The machine learning unit collects historical infant hip joint detection data, obtains a three-dimensional angle value of the relative inspection position coordinates of the ultrasound probe through machine learning, and transmits the three-dimensional angle value to the angle adjustment unit; wherein the historical infant hip joint detection data includes infant age, physiological parameters, body position parameters, ultrasound parameters and coupling agent parameters;

[0110] The probe orientation unit receives the probe image, identifies the probe reference point and the specific part of the hip joint according to medical common sense, and draws a line between the two to determine the probe ultrasound emission direction;

[0111] The angle adjustment unit receives the predicted three-dimensional angle value, and then after receiving the in-place instruction issued by the motion control submodule, uses the image registration technology to integrate the probe image and the inspection position into the image comparison system in the same coordinate space, establishes a three-dimensional coordinate system with the inspection position coordinates as the origin, and uses the vector analysis method to measure whether the angles between the probe's ultrasonic emission direction and the inspection position's XYZ coordinates meet the three-dimensional angle value. If not, an angle adjustment instruction is generated and sent to the motion control submodule to adjust the probe angle.

[0112] In a specific implementation, the dynamic human-machine coupling submodule includes a probe force detection unit, a data preprocessing unit, a shear thickening fluid controller unit and a stability analysis unit, wherein the probe force detection unit is connected to the data preprocessing unit, the data preprocessing unit is connected to the shear thickening fluid controller unit, and the shear thickening fluid controller unit is connected to the stability analysis unit;

[0113] The probe force detection unit measures the random force signal applied to the probe in real time by integrating a sensor located at the probe;

[0114] The data preprocessing unit filters, amplifies and digitizes the signals collected by the sensor to extract useful information;

[0115] The controller unit is used to design a dynamic characteristic model that can describe the interaction between the probe and the infant, convert the force signal into a control signal, and design a control algorithm to dynamically adjust the damping of the probe against stress according to the random force signal applied to the probe;

[0116] The controller unit includes a common admittance controller unit, a shear hardening glue controller unit and a shear thickening fluid controller unit. This application takes the shear thickening fluid controller unit as an example to describe in detail the process of designing a dynamic characteristic model of the shear thickening fluid that describes the interaction between the probe and the baby and designing a control algorithm;

[0117] The design process of the dynamic characteristic model is as follows:

[0118] Material property research, studying the rheological properties of shear thickening fluids, including the relationship between shear stress and shear rate; experimental data collection, collecting behavioral data of shear thickening fluid materials under different conditions through experimental methods; mathematical modeling, using the collected data to establish a mathematical model, usually using a power law model or other nonlinear models to describe the mechanical behavior of shear thickening fluid materials; computer simulation, using computational fluid dynamics (CFD) software to simulate and analyze the behavior of shear thickening fluid materials; model verification, verifying the accuracy and predictive ability of the model through experimental data; parameter identification, determining model parameters, such as yield stress, flow index, etc., to accurately describe the characteristics of shear thickening fluid materials; multi-physics coupling analysis, if necessary, perform multi-physics coupling analysis, such as fluid-structure interaction; model optimization, optimizing the model to improve prediction accuracy and computational efficiency.

[0119] The process of designing the control algorithm is as follows:

[0120] Requirements analysis: determine the control objectives and performance indicators, such as stability, accuracy, response time, etc.; System modeling: model the system, including the probe, the baby's motion model and the interaction between them; Control strategy selection: select the appropriate control strategy according to the system characteristics and requirements, such as PID, adaptive control, predictive control, etc.; Control algorithm design: design the detailed structure of the control algorithm, including control laws and parameters; Algorithm optimization: optimize the control algorithm according to the simulation results and adjust the parameters to meet the performance requirements; Hardware selection and integration: select the appropriate hardware platform and integrate the control algorithm into the hardware; Development of real-time system development: develop real-time control systems to ensure that the algorithm can run in a real-time environment; Hardware-in-the-loop testing: verify the integration and performance of the control algorithm with the physical system in a hardware-in-the-loop test environment; Stability and robustness analysis: perform stability and robustness analysis on the control algorithm to ensure that it can work reliably under various conditions; Safety feature integration: integrate safety features such as force limitation, speed limitation and emergency stop function; Field testing and adjustment: test the control algorithm in actual application scenarios and make adjustments based on the test results.

[0121] The stability analysis unit uses control theory and Lyapunov method to analyze the stability of the controller. Stability analysis can help predict the dynamic behavior of the shear thickening fluid system under different operating conditions, determine whether the system can reach a stable state or there is a risk of instability, and ensure that the system can quickly recover to a stable state when subjected to external disturbances. It can also improve the controller's response speed to external stress, reduce overshoot, and improve robustness.

[0122] In a specific implementation, the ultrasound probe adaptive adjustment submodule includes a key feature positioning unit, a standard section prediction unit, a probe angle and position measurement unit and a feature matching analysis unit, and the key feature positioning unit, the standard section prediction unit and the probe angle and position measurement unit are respectively connected to the feature matching analysis unit;

[0123] The key feature positioning unit is used to receive the ultrasound section image containing 8 anatomical landmarks and 3 identification points recognized in real time by the image recognition submodule, and locate the key features in the image using a convolutional neural network algorithm in the standard section image, the key features including the 8 anatomical landmarks and the 3 identification points;

[0124] The standard section prediction unit uses convolutional neural network training to obtain the standard section, calibrates the position coordinates of key features on the standard section in the XY coordinate system, and fits the connection lines between different key features. The standard section must include 8 anatomical landmarks and 3 identification points. The historical parameters collected during training include infant age, physiological parameters, body position parameters, ultrasound parameters, and coupling agent parameters. The model is tuned and the optimization algorithm is used to improve the accuracy of the model.

[0125] The probe angle and position measurement unit obtains the probe image in real time to measure the initial position and angle of the probe;

[0126] The feature matching and analysis unit acquires the real-time image and identifies multiple key features therein, and receives the standard section input by the user interaction submodule at the same time, integrates the trained standard section, real-time image and probe image into the image comparison system in the same coordinate space by using the image registration technology, acquires the same key features of the standard section and the real-time image, and compares and analyzes the positions and angles of the same key features of the above different images, calculates the difference between the position of the key feature in the real-time image and the position of the same key feature in the standard section, and then obtains the position error, calculates the difference between the angle of the probe and the angle of the standard section to obtain the angle error, and acquires the initial position and angle of the probe in the same coordinate space by using the image analysis technology according to the probe image, and sends the above position error, angle error and the initial position and angle information of the probe to the motion control submodule;

[0127] The motion control submodule is used to receive the control signal transmitted by the feature matching analysis unit, which contains angle error information, position error information, and the initial position and angle information of the probe. The target position and target angle information are calculated based on the angle error and position error information, and the initial position, initial angle, target position and target angle information are parsed into instructions for the movement of the robotic arm. The motion path of the robotic arm from the initial position to the target position and the rotation path from the initial angle to the target angle are planned, and the probe is driven from the current position to the target position and rotated by the corresponding angle through the robotic arm controller.

[0128] In a specific implementation, the feature matching analysis unit includes a position error calculation subunit and an angle error calculation subunit. The position error calculation subunit obtains the coordinates (XY plane) of a preset key feature from the real-time image. 当前 , Y 当前 ), obtain the coordinates of the same key feature on the XY plane from the predicted standard section (X 标准 , Y 标准 ), using direct comparison method: ΔX=(X 标准 -X 当前 ), ΔY=(Y 标准 -Y 当前 ), where (ΔX, ΔY) constitutes the position error;

[0129] The angle error calculation subunit uses a feature detection algorithm to obtain the positions of two preset key features from the real-time image, uses a straight line detection algorithm to obtain a line between the two preset key features, and defines the line as a vector A, and determines the direction of the vector;

[0130] Using a feature detection algorithm to obtain the positions of the same two preset key features in the standard section from the predicted standard section, using a straight line detection algorithm to obtain a connecting line between the two preset key features, and defining the connecting line as a vector B, and determining the direction of the vector;

[0131] Using the formula Calculate the angle θ between the two vectors A and B, where · represents the dot product and ∥·∥ represents the modulus of the vectors. The angle θ constitutes the angle error.

[0132] In specific implementation, the image analysis module includes an image recognition submodule, a key point positioning submodule, a line extraction submodule, an angle calculation submodule and a typing diagnosis submodule, the image recognition submodule is connected to the identification point positioning unit, the identification point positioning unit is connected to the line extraction submodule, the line extraction submodule is connected to the angle calculation submodule, and the angle calculation submodule is connected to the typing diagnosis submodule;

[0133] The image recognition submodule receives the standard section containing 8 anatomical landmarks and 3 identification points captured by the ultrasound probe adaptive adjustment submodule, and uses machine learning technology to identify the cartilage roof, acetabular roof, the lower edge of the ilium in the deep part of the acetabular fossa, the acetabular labrum and the middle interface of the acetabular roof, for example, the above-mentioned medical anatomical landmarks are identified by an automatic image recognition module;

[0134] The key point positioning submodule uses a feature detection algorithm to enhance the edge and locate the position of the key point from the hip joint image identified by the image recognition submodule;

[0135] The line extraction submodule uses a line detection algorithm to extract a straight line from the image based on the middle section of the acetabulum top, the position of the key points and the straight line definition;

[0136] The angle calculation submodule converts the lines into vectors and uses trigonometric functions to calculate the angles between them, wherein the angle between the baseline and the bone top line is defined as angle α, and the baseline intersects with the cartilage top line to form angle β;

[0137] The classification diagnosis submodule performs threshold judgment on the above angles and outputs corresponding classification results, wherein the threshold classification includes Graf I type: α angle ≥ 60°; Graf II type: 50°≤α<60°, age less than 3 months is type II a, age greater than 3 months is type II b; Graf IIc type: 43°≤a<50°, β≤77°; Graf D type: 43°≤a<50°, β>77°; Graf IIIa type: a<43°, the cartilage top is pushed and compressed toward the head end, and the cartilage top has no degeneration; Graf IIIb type: a<43°, the cartilage top is pushed and compressed toward the head end, the cartilage top is compressed and degenerated, and the echo is enhanced; Graf IV type: a<43°, the cartilage top is pushed and compressed toward the foot end. Doctors further refine the classification based on the above angle threshold combined with the infant's age, cartilage roof imaging, etc. For example, Graf III type can be further divided into Graf IIIa type and Graf IIIb type based on whether the cartilage roof is degenerated.

[0138] Wherein, the key point positioning submodule includes an edge enhancement subunit and a corner point detection subunit;

[0139] The edge enhancement subunit uses an edge detection algorithm to highlight the lower edge of the ilium, the acetabular labrum and the middle interface of the acetabular bone top in the deep part of the acetabular fossa in the image;

[0140] The corner point detection subunit uses Harris corner point detection technology to identify key points in the edge of the image, and the key points include the top of the acetabulum, the lower edge of the iliac ramus of the acetabulum bottom, the acetabulum bone edge and the center of the acetabular labrum.

[0141] In specific implementation, the line extraction submodule includes a line definition unit, a candidate line detection unit and an optimal line determination unit;

[0142] The straight line definition unit is used to define a straight line, wherein the base line is defined as the tangent line of the ilium from the point where the proximal perichondrium changes to the periosteum. First, the uppermost end of the hyaline cartilage top is found. At this point, the proximal perichondrium changes to the periosteum. From this fulcrum, the tangent line of the ilium is drawn caudally; the bone top line is defined as the tangent line of the bone top from the lower edge of the ilium; the cartilage top line is defined as the line connecting the bone edge to the midpoint of the labrum;

[0143] The candidate line detection unit maps the edge points and key points of the interface in the middle of the ilium to the parameter space through Hough transform according to the definition of the line, and detects the line through the accumulator method to form a candidate line;

[0144] The best straight line determination unit uses the RANSAC algorithm to randomly select a group of points from edge points or key points for straight line fitting, and evaluates the consistency of other points with the straight line, and finally determines the best straight line.

[0145] The present invention and its embodiments are described above, which is not restrictive. What is shown in the accompanying drawings is only one of the embodiments of the present invention, and the actual content is not limited to this. In short, if ordinary technicians in this field are inspired by it and do not deviate from the purpose of the invention, they can creatively design structural methods and embodiments similar to the technical scheme, which should all fall within the scope of protection of the present invention.

Claims

1. A robot system for adaptively adjusting ultrasound imaging probes for standard position of infant hip joints, characterized in that: It includes an ultrasonic probe module, a mechanical arm artificial intelligence module, an image automatic recognition module and an image analysis module, wherein the ultrasonic probe module is connected to the mechanical arm artificial intelligence module and the image automatic recognition module, the mechanical arm artificial intelligence module is connected to the image automatic recognition module and the image analysis module, and the ultrasonic probe module and the image automatic recognition module are respectively connected to the image analysis module; The ultrasound probe module is responsible for controlling the probe to capture ultrasound images of the infant's hip joint, including controlling the emission and pause of corresponding ultrasound waves according to preset values, and is responsible for acquiring probe images in real time through sensors or cameras; The robotic arm artificial intelligence module is responsible for controlling the robotic arm, which further controls the precise movement, positioning and damping of random stress of the probe; The automatic image recognition module uses machine learning technology to automatically recognize ultrasound sections that meet medical requirements in ultrasound images, thereby reducing the manual operation of doctors. The image analysis module applies image recognition technology to simulate the expert diagnosis process and realize intelligent image analysis and diagnosis.

2. The infant hip joint standard position ultrasound imaging probe adaptive adjustment robot system as claimed in claim 1, characterized in that: The image automatic recognition module includes an image collection submodule, an image preprocessing submodule, a machine modeling submodule and a model optimization submodule, wherein the image collection submodule is connected to the image preprocessing submodule, the image preprocessing submodule is connected to the machine modeling submodule, and the machine modeling submodule is connected to the model optimization submodule; The image collection submodule is used to obtain a large number of infant hip joint ultrasound images. These images should include hip joint images of different angles and different infants. Professional doctors mark the key features of the hip joint on the images, including 8 anatomical landmarks and 3 identification points. The image preprocessing submodule is used to remove noise and artifacts from the image, improve image quality, adjust the size and contrast of the image to make it suitable for subsequent processing, and enhance the key features annotated in the image; The machine modeling submodule uses a convolutional neural network for model training, including using the convolutional neural network to automatically extract key features in the image, dividing the historical ultrasound images into a training set and a test set, training the recognition prediction model with the training set, and tuning the model with the test set. The recognition prediction model is used to identify ultrasound sections containing 8 anatomical landmarks and 3 identification points; The model optimization submodule defines the cross entropy loss function, which is used to measure the difference between the model prediction and the actual annotation. The SGD optimization algorithm is selected to train the network and find the minimum value of the loss function, thereby training a performance-optimized model.

3. The standard position probe adaptive robot system for infant ultrasonic hip joint imaging as claimed in claim 2, characterized in that: The robot arm artificial intelligence module also includes a path planning submodule, a motion control submodule, an angle control submodule, a tactile perception submodule, a coupling agent automatic replacement and cleaning submodule, a dynamic human-machine coupling submodule, an ultrasonic probe adaptive adjustment submodule and a user interaction submodule, wherein the path planning submodule is respectively connected to the motion control submodule and the coupling agent automatic replacement and cleaning submodule, the motion control submodule is respectively connected to the tactile perception submodule, the angle control submodule, the dynamic human-machine coupling submodule and the ultrasonic probe adaptive adjustment submodule, and the path planning submodule, the motion control submodule, the coupling agent automatic replacement and cleaning submodule, the angle control submodule, the dynamic human-machine coupling submodule and the ultrasonic probe adaptive adjustment submodule are respectively connected to the user interaction submodule; Path planning submodule: responsible for calculating the optimal path for the ultrasound probe to reach the baby's examination site, ensuring that the probe can reach the target position quickly and accurately, and controlling the start and stop of the robotic arm; Motion control submodule: The motion planning algorithm is used to convert the path planning information provided by the path planning submodule, the dynamic path adjustment unit, the angle adjustment unit and the feature matching analysis unit into specific motion instructions for each joint of the robotic arm. The motion controller controls the movement and operation of the robotic arm to ensure that the ultrasound probe can move smoothly and accurately to the predetermined position according to the planned path. Tactile perception submodule: Through the feedback of force tactile array sensor, proximity array sensor and joint torque sensor, the robot arm is controlled to simulate the doctor's arm, and the contact area and contact force between the probe and the baby's skin are adjusted to achieve the best optimization; The angle control submodule uses a machine learning algorithm to predict the initial angle of the ultrasonic probe at the inspection position, and after receiving the position arrival instruction, adjusts the angle value of the probe at the inspection position to meet the angle setting requirements; Coupling agent automatic replacement and cleaning submodule: realizes automatic replacement of disposable coupling agent before examination, and automatic cleaning and sterilization of the probe and examination site after examination; Dynamic human-machine coupling submodule: adjusts the probe's resistance to the stress from the random movements of the baby, solving the problem of the baby's random movements interfering with the probe's imaging quality; Ultrasonic probe adaptive adjustment submodule: According to the positional relationship between the detected image and the predicted standard section, the position and angle of the ultrasonic probe are automatically adjusted to accurately capture the standard section; User interaction submodule: allows doctors or operators to interact with the robotic arm, including data input and necessary manual adjustments and interventions.

4. The standard position probe adaptive robot system for infant ultrasonic hip joint imaging as claimed in claim 3, characterized in that: The path planning submodule includes an environment perception unit, a map reconstruction unit, a target recognition and positioning unit, a path search algorithm unit, a dynamic path adjustment unit and a user intervention interface unit, wherein the environment perception unit is connected to the map reconstruction unit, the target recognition and positioning unit is connected to the path search algorithm unit, and the map reconstruction unit, the dynamic path adjustment unit and the user intervention interface unit are all connected to the path search algorithm unit; The environment sensing unit collects information about the examination environment, including the position of the infant, the specific location of the infant's examination part, and the layout of the examination bed; The map reconstruction unit constructs or updates a map of the inspection area for path planning based on data provided by the environment perception submodule; The target recognition and positioning unit uses a convolutional neural network to recognize the specific location of the infant's examination site, including the greater trochanter of the femur at the hip, and uses a spatial transformation technology to convert the hip joint image coordinates into actual physical space coordinates to provide accurate examination site coordinate information, while obtaining the current position coordinates of the probe through a position detection sensor; The path search algorithm unit uses a path search algorithm to calculate an optimal path from a current position to a target position, and the algorithm includes an A* algorithm, a Dijkstra algorithm or an RRT algorithm; The dynamic path adjustment unit monitors environmental changes and infant movements in real time during path execution, and dynamically adjusts the path to adapt to new situations, including using sensors and cameras to monitor the environment and infant movements in real time, using computer vision technology to track the movement of the infant and the examination site, analyzing movement patterns and predicting future positions, and applying a path adjustment algorithm to recalculate or adjust the path; The user intervention interface unit allows a doctor or operator to manually intervene in the path planning process when necessary to ensure safety and examination quality.

5. The infant hip joint standard position ultrasound imaging probe adaptive adjustment robot system as claimed in claim 4, characterized in that: The angle control submodule includes a machine learning unit, a probe orientation unit and an angle adjustment unit, and the machine learning unit and the probe orientation unit are respectively connected to the angle adjustment unit; The machine learning unit collects historical infant hip joint detection data, obtains a three-dimensional angle value of the relative inspection position coordinates of the ultrasound probe through machine learning, and transmits the three-dimensional angle value to the angle adjustment unit; wherein the historical infant hip joint detection data includes infant age, physiological parameters, body position parameters, ultrasound parameters and coupling agent parameters; The probe orientation unit receives the probe image, identifies the probe reference point and the specific part of the hip joint according to medical common sense, and draws a line between the two to determine the probe ultrasound emission direction; The angle adjustment unit receives the predicted three-dimensional angle value, and then after receiving the in-place instruction issued by the motion control submodule, uses the image registration technology to integrate the probe image and the inspection position into the image comparison system in the same coordinate space, establishes a three-dimensional coordinate system with the inspection position coordinates as the origin, and uses the vector analysis method to measure whether the angles between the probe's ultrasonic emission direction and the inspection position's XYZ coordinates meet the three-dimensional angle value. If not, an angle adjustment instruction is generated and sent to the motion control submodule to adjust the probe angle.

6. The infant hip joint standard position ultrasound imaging probe adaptive adjustment robot system as claimed in claim 3, characterized in that: The dynamic human-machine coupling submodule includes a probe force detection unit, a data preprocessing unit, a controller unit and a stability analysis unit, wherein the probe force detection unit is connected to the data preprocessing unit, the data preprocessing unit is connected to the controller unit, and the controller unit is connected to the stability analysis unit; The probe force detection unit measures the random force signal applied to the probe in real time by integrating a sensor located at the probe; The data preprocessing unit filters, amplifies and digitizes the signals collected by the sensor to extract useful information; The controller unit is used to design a dynamic characteristic model that can describe the interaction between the probe and the infant, convert the force signal into a control signal, and design a control algorithm to dynamically adjust the damping of the probe against stress according to the random force signal applied to the probe; The stability analysis unit uses control theory and Lyapunov method to analyze the stability of the controller.

7. The infant hip joint standard position ultrasound imaging probe adaptive adjustment robot system as claimed in claim 5, characterized in that: The ultrasound probe adaptive adjustment submodule includes a key feature positioning unit, a standard section prediction unit, a probe angle and position measurement unit and a feature matching analysis unit, and the key feature positioning unit, the standard section prediction unit and the probe angle and position measurement unit are respectively connected to the feature matching analysis unit; The key feature positioning unit is used to receive the ultrasound section image containing 8 anatomical landmarks and 3 identification points recognized in real time by the image recognition submodule, and locate the key features in the image using a convolutional neural network algorithm in the standard section image, the key features including the 8 anatomical landmarks and the 3 identification points; The standard section prediction unit uses convolutional neural network training to obtain the standard section, calibrates the position coordinates of key features on the standard section in the XY coordinate system, and fits the connection lines between different key features. The standard section must include 8 anatomical landmarks and 3 identification points. The historical parameters collected during training include infant age, physiological parameters, body position parameters, ultrasound parameters, and coupling agent parameters. The model is tuned and the optimization algorithm is used to improve the accuracy of the model. The probe angle and position measurement unit obtains the probe image in real time to measure the initial position and angle of the probe; The feature matching and analysis unit acquires the real-time image and identifies multiple key features therein, and receives the standard section input by the user interaction submodule at the same time, integrates the trained standard section, real-time image and probe image into the image comparison system in the same coordinate space by using the image registration technology, acquires the same key features of the standard section and the real-time image, and compares and analyzes the positions and angles of the same key features of the above different images, calculates the difference between the position of the key feature in the real-time image and the position of the same key feature in the standard section, and then obtains the position error, calculates the difference between the angle of the probe and the angle of the standard section to obtain the angle error, and acquires the initial position and angle of the probe in the same coordinate space by using the image analysis technology according to the probe image, and sends the above position error, angle error and the initial position and angle information of the probe to the motion control submodule; The motion control submodule is used to receive the control signal transmitted by the feature matching analysis unit, which contains angle error information, position error information, and the initial position and angle information of the probe. The target position and target angle information are calculated based on the angle error and position error information, and the initial position, initial angle, target position and target angle information are parsed into instructions for the movement of the robotic arm. The motion path of the robotic arm from the initial position to the target position and the rotation path from the initial angle to the target angle are planned, and the probe is driven from the current position to the target position and rotated by the corresponding angle through the robotic arm controller.

8. The infant hip joint standard position ultrasound imaging probe adaptive adjustment robot system as claimed in claim 7, characterized in that: The feature matching analysis unit includes a position error calculation subunit and an angle error calculation subunit. The position error calculation subunit obtains the coordinates (XY plane) of a preset key feature from the real-time image. 当前 , Y 当前 ), obtain the coordinates of the same key feature on the XY plane from the predicted standard section (X 标准 , Y 标准 ), using direct comparison method: ΔX=(X 标准 -X 当前 ), ΔY=(Y 标准 -Y 当前 ), where (ΔX, ΔY) constitutes the position error; The angle error calculation subunit uses a feature detection algorithm to obtain the positions of two preset key features from the real-time image, uses a straight line detection algorithm to obtain a line between the two preset key features, and defines the line as a vector A, and determines the direction of the vector; Using a feature detection algorithm to obtain the positions of the same two preset key features from the predicted standard section, using a straight line detection algorithm to obtain a line between the two preset key features, and defining the line as a vector B, and determining the direction of the vector; Using the formula Calculate the angle θ between the two vectors A and B, where · represents the dot product and ∥·∥ represents the modulus of the vectors. The angle θ constitutes the angle error.

9. The infant hip joint standard position ultrasound imaging probe adaptive adjustment robot system as claimed in claim 8, characterized in that: The image analysis module includes an image recognition submodule, a key point positioning submodule, a line extraction submodule, an angle calculation submodule and a typing diagnosis submodule, wherein the image recognition submodule is connected to the identification point positioning unit, the identification point positioning unit is connected to the line extraction submodule, the line extraction submodule is connected to the angle calculation submodule, and the angle calculation submodule is connected to the typing diagnosis submodule; The image recognition submodule receives the standard section containing 8 anatomical landmarks and 3 identification points captured by the ultrasound probe adaptive adjustment submodule, and uses machine learning technology to identify the cartilage roof, acetabular bone roof, the lower edge of the ilium in the deep part of the acetabular fossa, the acetabular labrum and the middle section of the acetabular bone roof; The key point positioning submodule uses a feature detection algorithm to enhance the edge and locate the position of the key point from the hip joint image identified by the image recognition submodule; The line extraction submodule uses a line detection algorithm to extract a straight line from the image based on the middle section of the acetabulum top, the position of the key points and the straight line definition; The angle calculation submodule converts the lines into vectors and uses trigonometric functions to calculate the angles between them, wherein the angle between the baseline and the bone top line is defined as angle α, and the baseline intersects with the cartilage top line to form angle β; The typing diagnosis submodule performs threshold judgment on the above angles and outputs corresponding typing results, wherein the threshold typing includes Graf I type: α angle ≥ 60°; Graf II type: 50°≤α<60°; Graf IIc type: 43°≤a<50°, β≤77°; Graf D type: 43°≤a<50°, β>77°; Graf IIIa type: a<43°; Graf IIIb type: a<43°; Graf IV type: a<43°; Wherein, the key point positioning submodule includes an edge enhancement subunit and a corner point detection subunit; The edge enhancement subunit uses an edge detection algorithm to highlight the lower edge of the ilium, the acetabular labrum and the middle section of the acetabular bone top in the deep part of the acetabular fossa in the image; The corner point detection subunit uses Harris corner point detection technology to identify key points in the edge of the image, and the key points include the top of the acetabulum, the lower edge of the ilium in the deep part of the acetabulum, the acetabulum bone edge and the midpoint of the acetabular labrum.

10. The infant hip joint standard position ultrasound imaging probe adaptive adjustment robot system as claimed in claim 9, characterized in that: The line extraction submodule includes a line definition unit, a candidate line detection unit and an optimal line determination unit; The straight line definition unit is used to define a straight line, wherein the baseline is defined as a tangent line of the ilium along the transition from the proximal perichondrium to the periosteum; the bone top line is defined as a tangent line of the acetabular bone top from the lower edge of the ilium in the deep part of the acetabular fossa; the cartilage top line is defined as a line connecting the bone edge to the midpoint of the labrum; The candidate line detection unit maps the edge points and key points of the interface in the middle of the ilium to the parameter space through Hough transform according to the definition of the line, and detects the line through the accumulator method to form a candidate line; The best straight line determination unit uses the RANSAC algorithm to randomly select a group of points from the above edge points or key points for straight line fitting, and evaluates the consistency of other points with the straight line, and finally determines the best straight line.

Images